Enable javascript in your browser for better experience. Need to know to enable it? Go here.
Financial services firm

Beyond the mainframe: Transforming legacy insurance systems with agentic AI

Our client is a prominent financial services firm headquartered in the US. They specialize in life insurance, investment management and policy administration. With approximately $12 billion in annual revenue, they manage highly sensitive  contract data for customers at a massive scale.

Navigating 40 years of technical debt

 

The organization relied on a core policy administration system to manage contracts, transaction details and other essential operations. This system contained four million lines of legacy code written in COBOL and IBM High Level Assembler (HLASM), two aging programming languages commonly found in long-standing financial environments. COBOL typically handles business processing, while HLASM is used for low-level, performance-critical tasks. 

 

The architecture presented an added layer of complexity because the system featured COBOL programs calling HLASM code and HLASM modules calling back into COBOL, resulting in tightly coupled dependencies that made modernization even more challenging. The internal team responsible for maintaining this intricate codebase was nearing retirement, which created an urgent need for modernization. Initial in-house efforts to update the system proved slow, unproductive and unsustainable.

Why Thoughtworks: Proven methodologies for high-stakes modernization

 

The financial services firm actively sought a solution and discovered Thoughtworks through online research. They read our published white papers detailing the capabilities of Code to Spec for reverse engineering legacy systems. Impressed by these documented approaches, the client approached us to prove our tools could successfully modernize their complex code.

Legacy modernization powered by AI/works™ 

 

To address their modernization goals, the client requested that we start with a proof of concept (POC). The POC centered on the beneficiary change functionality, a core part of their life insurance policy system. The client initially provided 20,000 lines of HLASM code for this key process, but due to intricate system dependencies, the scope quickly expanded to 160,000 lines of intertwined HLASM and COBOL code.

 

AI/works™, Thoughtworks' agentic development platform, led the modernization of the extended warranty workload by applying our engineering effectiveness approach to tackle this complex web of legacy languages. Using Code to Spec — a capability within AI/works™ — we reverse engineered the code and generated detailed functional specifications, iteratively validated with client-side subject matter experts.


We also extended Code to Spec to support HLASM reverse engineering, a capability that did not exist prior to this engagement. Finally, we used Claude Code for forward engineering to come up with microservices, which included forward engineering mainframe screens built with assembler, COBOL and DB2 to new services built on React, Java Spring Boot and Postgres as part of the same POC.

From months to weeks: A blueprint for large-scale digital transformation

 

  • We delivered the project in just four weeks, compared to the originally planned eight weeks.

 

  • The solution achieved 80% accuracy in code comprehension, showing dramatically improved productivity over their previous manual efforts.

 

The client was pleased with the accuracy, speed and overall engineering effectiveness of the automated process. The project paved the way for other POCs, leading to the successful integration of the HLASM processing capability back into Code to Spec to be used for future customers.

 

Explore our AI/works™ platform